Hugging Face's logo

TheAiCollectiveART
/
Glyph-Stream-31b

gemma4
language-u
gemma-4
glyph-stream
octonion-s7
compressed-tensors
Model card Files
xet
 Community

Configuration Parsing Warning:In config.json: "num_experts" must be a number

Configuration Parsing Warning:In config.json: "num_experts_per_tok" must be a number

Glyph-Stream-31b (TheAiCollectiveART/Glyph-Stream-31b)

Public repository of the 256-weight Language-U Gemma 4 31B (gemma-4-31b-it-qat-w4a16-ct) Dual-Road Sandwich glyph stream and companion runtime tables.

Verified Contents

  • gemma4_31b_language_u_sandwich_stream.bin (7,080,837,136 B / 7.08 GB dec / 6.59 GiB): Dual-Road Sandwich Stream (18,923,520 Opt4 132 B boundary glyphs + 95,477,760 Opt5 48 B middle glyphs behind 16 B LANU header; 114,401,280 total glyphs = 29,286,727,680 linear weights, R = 0).
  • gemma4_31b_embed_tokens.bin (2,818,572,288 B): Companion token embedding table (262,144 x 5,376 16-bit).
  • gemma4_31b_norms.safetensors (2,990,082 B): Layer norm weights.
  • s7_codebook_2048x8.bin (65,536 B): 2,048-vertex unit octonion codebook on S^7.
  • gemma4_31b_tensor_layout.json, manifest_gemma4_31b.json, config.json, generation_config.json, tokenizer.json, tokenizer_config.json, chat_template.jinja.
  • OFFICIAL_SAFETENSORS_LANGUAGE_U_VERIFIED_REPORT.json, PHYSICAL_PI4_31B_VERIFIED_REPORT.json.

Code & verification suite: https://github.com/DannyB-bit/Language-U-Gemma4

Downloads last month
44
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support